Safety and governance
Data provenance
What is data provenance?
Data provenance is the record of where each training example came from, when, and under what permission. It is what allows a specific model behaviour to be traced back to its cause, and what allows a licensing question to be answered without guesswork.
Provenance becomes urgent at exactly the wrong moment: when a customer asks whether their data was used, when a model produces something it should not have, or when a licence question arrives from a lawyer. None of those are answerable retrospectively without records kept in advance.
The practical form is a per-record trail — source, timestamp, transformation applied, and whether the record was generated or collected — carried alongside the dataset rather than in someone’s memory.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.